[
https://issues.apache.org/jira/browse/FLINK-9794?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16541068#comment-16541068
]
ASF GitHub Bot commented on FLINK-9794:
---------------------------------------
Github user jrthe42 commented on the issue:
https://github.com/apache/flink/pull/6301
Hi @sihuazhou , I am not familiar with the checkpoint mechanism of Flink,
and I check the source code again.
Although the ```RichSinkFunction#invoke()``` and
```RichSinkFunction#snapshotState()``` are not executed in the same thread, but
there is already synchronization mechanism in ```StreamTask```.
```StreamTask``` use a **checkpoint lock object** to make sure they won't be
called concurrently. Check ```StreamTask#performCheckpoint()``` and
```StreamInputProcessor#processInput()``` if you want to know more.
Thanks for your comment, I removed synchronization here, and this PR is
updated. cc @yanghua
> JDBCOutputFormat does not consider idle connection and multithreads
> synchronization
> -----------------------------------------------------------------------------------
>
> Key: FLINK-9794
> URL: https://issues.apache.org/jira/browse/FLINK-9794
> Project: Flink
> Issue Type: Bug
> Components: Streaming Connectors
> Affects Versions: 1.4.0, 1.5.0
> Reporter: wangsan
> Priority: Major
> Labels: pull-request-available
>
> Current implementation of JDBCOutputFormat has two potential problems:
> 1. The Connection was established when JDBCOutputFormat is opened, and will
> be used all the time. But if this connection lies idle for a long time, the
> database will force close the connection, thus errors may occur.
> 2. The flush() method is called when batchCount exceeds the threshold, but it
> is also called while snapshotting state. So two threads may modify upload and
> batchCount, but without synchronization.
> We need fix these two problems to make JDBCOutputFormat more reliable.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
